Text to speech
Paste some text, hear it read aloud in seconds
Free to listen, no account needed; generation and export are billed by compute, with an estimate up front
The hard part of turning text into audio isn't whether the machine sounds human — models cleared that bar a while ago. It's the words it gets wrong: homographs, numbers, proper nouns. Three slips in a 200-word script and the whole thing sounds “machine-read”.
So proofreading comes first: anything likely to be mispronounced is flagged for you to confirm before a single word is spoken. Confirmed pronunciations go into this project's glossary and apply to everything after — you don't correct the same word twice.
What it does
Several voices that don't collide
When several people speak in one passage, pitch and timbre are laid out apart. Pitch alone isn't enough — we measured two characters 22 Hz apart whose timbre was identical, and it still sounded like one person talking to themselves.
Emotion is a fixed set, not free text
Ten levels: neutral, tense, sad, whisper and so on. Write something like “wistful with a hint of hope” and the engine quietly falls back to neutral — it won't tell you, so you think it's acting while it's just reading.
Proofread first, speak second
Homographs, number readings and proper nouns are surfaced for you to confirm. TTS will read a typo perfectly fluently — skip this step and you won't catch it even after listening to the whole thing.
Seconds for short text, a queue for long
A 200-word script is a few seconds' wait. A whole book isn't — it goes into a job queue, you can close the page, and we'll tell you when it's done.
How it works
-
Paste it in
Ctrl+V straight onto the page, or drop a .txt file. No sign-up first.
-
Glance at the proofreading
Likely mispronunciations are flagged. What you confirm goes into the glossary and applies from then on.
-
Pick a voice, listen
Swap the voice and listen again, or adjust pace and emotion. Export when you're happy.
FAQ
- Is it free?
- Listening is free and needs no account. Export and long-form generation are billed by the compute actually used — you'll see the estimate before anything starts.
- Is my text stored?
- Not while you're just listening. Once you create a project it lives in your own account and you can delete it any time. Generation runs on our servers, so at that step the text does leave your browser — we're not going to be vague about that.
- How long can the text be?
- Up to 2,000 characters per preview — that's a reasonable ceiling for “standing there waiting”. Longer material goes through the project flow: a full-length book is split into episodes and generated line by line, so a failure only reruns that one line.
- Which languages are supported?
- The interface is in 16 languages. What gets read aloud depends on the voice you pick — Chinese and English have the fullest sets, others are being filled in.
Text to speech
Free to listen, no account needed; generation and export are billed by compute, with an estimate up front